Monthly Total Count from Multiple Tables PostgreSQL - sql

Question

I have a search set up which gives total count of new patient visits and total count of patient visits, and comparing the totals for the requested year to the previous year's totals.
The SQL queries the date fields firstexam and lastexam from the table patient_info.
I have since found out that some users do not update the lastexam with every patient visit, and therefore the lastexam would not give the total number of patient visits.
Total number of patient visits can be obtained by searching the transactions table. Invoices in the transaction table are marked with the column transtype as 'Inv'. So, the total number of patient visits would be the total number of invoices in the date range (taking into account that two invoices entered for a patient in a single day count as one visit).
Below is the code for the SQL query set up based on firstexam and lastexam.
I have been struggling with this for some time now and am stuck. Any help would be greatly appreciated.
select
to_char(('2012-' || m || '-01')::date, 'Month'),
thisyear, lastyear, totalthisyear, totallastyear
from (
select
extract(month from m) as m,
sum(case
when firstexam between '2013-01-01' and '2013-12-31' then firstexam_count
else 0 end
) as thisyear,
sum(case
when firstexam between '2012-01-01' and '2012-12-31' then firstexam_count
else 0 end
) as lastyear,
sum(case
when lastexam between '2013-01-01' and '2013-12-31' then lastexam_count
else 0 end
) as totalthisyear,
sum(case
when lastexam between '2012-01-01' and '2012-12-31' then lastexam_count
else 0 end
) as totallastyear
from
generate_series (
'2012-01-01'::date, '2013-12-31', '1 month'
) g(m)
left join (
select count(*) as firstexam_count, date_trunc('month', firstexam) as firstexam
from patient_info
where firstexam between '2012-01-01' and '2013-12-31'
group by 2
) pif on firstexam = m
left join (
select count(*) as lastexam_count, date_trunc('month', lastexam) as lastexam
from patient_info
where lastexam between '2012-01-01' and '2013-12-31'
group by 2
) pil on lastexam = m
group by 1
) s
order by m

Answer 1

If you want to report information about exams, you ought to store information about exams.
More specifically, if you want to count exams, you ought to store information about each exam.
Don't use column names like "thisyear" and "lastyear". This year isn't 2013, although that's how you present it.
Usually, visits and exams are different things. Be careful with terminology. (Here it's not such a big deal, because we don't have information about either visits or exams. Only about invoices. Still, it's a good habit.)
If you're concerned about a particular output format, ask yourself whether you're building a query or a report. Build queries in SQL. Build reports with a report writer or application code.
For simplicity, I'm going to
ignore the "patient_info" table,
ignore the outer join you need in order to generate zeroes for months in which there were no exams, and
use common table expressions. (In production I'd rather use views than common table expressions).
Let's start with just a table of transactions.
create table transactions (
ptnumber INT,
dateofservice date,
transtype varchar(3)
);
-- Not quite the same data you started with.
insert into transactions (ptnumber, dateofservice, transtype)
values
(1, '2012-01-01', 'Inv'),
(1, '2012-02-11', 'Inv'),
(2, '2012-01-02', 'Inv'),
(3, '2013-01-01', 'Inv'),
(4, '2013-02-12', 'Inv'),
(5, '2012-12-31', 'Inv'),
(5, '2013-12-31', 'Inv'),
(5, '2013-12-31', 'Inv'),
(6, '2013-06-21', 'Inv');
You said "two invoices entered for a patient in a single day count as one [exam]". I guess that means two or more. So we can extract the set of patient exams like this. I expect two rows for patient 5--one in 2012 and one in 2013.
select distinct ptnumber, dateofservice
from transactions
where transtype = 'Inv'
and dateofservice between '2012-01-01' and '2013-12-31'
order by ptnumber;
ptnumber dateofservice
--
1 2012-01-01
1 2012-02-11
2 2012-01-02
3 2013-01-01
4 2013-02-12
5 2012-12-31
5 2013-12-31
6 2013-06-21
This is the key to your whole problem--a set of distinct patient exams over a defined range of dates. Based on this set, counting patient visits by month is straightforward. (Counting them every which way is straightforward.)
with patient_exams as (
select distinct ptnumber, dateofservice
from transactions
where transtype = 'Inv'
and dateofservice between '2012-01-01' and '2013-12-31'
)
select to_char(dateofservice, 'YYYY-MM') as month_of_service, count(*) as num_patient_exams
from patient_exams
group by 1
order by 1;
month_of_service num_patient_visits
--
2012-01 2
2012-02 1
2012-12 1
2013-01 1
2013-02 1
2013-06 1
2013-12 1
First exams
Again, start by deriving a set that will give you reliable counts. You want one row per patient, and you want the earliest invoice date. The date of a patient's first exam has nothing to do with the date range you want to report; including the date range in this query's WHERE clause will give you the wrong data.
select ptnumber, min(dateofservice) as first_exam_date
from transactions
where transtype = 'Inv'
group by ptnumber
order by ptnumber;
ptnumber first_exam_date
--
1 2012-01-01
2 2012-01-02
3 2013-01-01
4 2013-02-12
5 2012-12-31
6 2013-06-21
Now counting how many new patients you gained each month is straightforward.
with first_exams as (
select ptnumber, min(dateofservice) as first_exam_date
from transactions
where transtype = 'Inv'
group by ptnumber
)
select to_char(first_exam_date, 'YYYY-MM') exam_month, count(*) num_first_exams
from first_exams
where first_exam_date between '2012-01-01' and '2013-12-31'
group by 1
order by 1;
exam_month num_first_exams
--
2012-01 2
2012-12 1
2013-01 1
2013-02 1
2013-06 1